Method for angle calibration of the position of a video camera on board an automotive vehicle

Abstract

The method consists in taking, when the vehicle is running, images of the road markings delimiting the circulation lanes of the road, and in fully estimating, through an iterative process, the orientation of the camera with respect to the vehicle, based on the position of two lanes located side-by-side on the image. The calibration essentially comprises: correcting the position of the lane edges in the image ( 10, 12 ); estimating the residual pitch and yaw ( 16 ); updating the rotation matrix ( 18 ); estimating the residual roll ( 20 ); updating the rotation matrix ( 24 ). These steps are iterated until the corrective angles estimated by each module are negligible ( 22 ).

The invention claimed is: 
     
       1. A method of angle calibration of the position of a video camera on board an automotive vehicle, by estimation of a calibration matrix (R e ) defining a rotation between a coordinate system linked to the camera (X c , Y c , Z c ) and a coordinate system linked to the vehicle (X v , Y v , Z v ), this rotation being defined as pitch (a x ), yaw (a y ) and roll (a z ) angles,
 the method comprising, during a phase of displacement of the vehicle on a road, the steps of: 
 a) taking a sequence of video images of a scene of the road, each image of the sequence comprising a plurality of traces of interest representative of ground markings of parallels circulation lanes; 
 b) detecting edges ( 10 ) on the video images, to obtain a list of rectilinear edges (D k ) corresponding, in each image, to said traces of interest; 
 c) correcting ( 12 ) the list of edges by applying to each edge a rotation according to the estimate of the calibration matrix; 
 d) calculating ( 16 ) residual pitch and yaw angles by a method of detection of the vanishing point applied to the corrected list of edges obtained at step c); 
 e) updating ( 18 ) the estimate of the calibration matrix by applying a rotation according to the residual pitch and yaw angles calculated at step d); 
 the method being characterized in that it comprises the following successive steps: 
 f) calculating ( 20 ;  28 ,  32 ,  34 ,  36 ) a residual roll angle from the corrected list of edges obtained at step c) and the estimate of the calibration matrix obtained at step e), by:
 estimating the respective widths of at least two circulation lanes based on the list of edges; 
 comparing the so-estimated respective widths and calculating at least one lane width difference; and 
 calculating the roll angle as a function the so-calculated width difference, the real widths of the lanes being supposed to be identical; 
 
 g) updating ( 40 ) the estimate of the calibration matrix by applying a rotation according to the residual roll angle calculated at step f); and 
 h) iterating, on at least one sequence of images, steps c) to g) until the residual pitch, yaw and roll angles are lower than a predetermined threshold. 
 
     
     
       2. The method of  claim 1 , wherein the step f) of estimating the roll angle comprises the following successive steps:
 f1) correcting ( 28 ) the list of edges by applying to each edge a rotation according to the estimate of the calibration matrix and a rotation according to the estimate of a roll matrix; 
 f2) for each image, based on the list of edges obtained at step f1) and as a function of a lane width reference (W e ):
 calculating ( 32 ) the width of each of the circulation lanes detected in the image, and 
 calculating ( 32 ) a residual roll angle for each couple of circulation lanes detected in the image; 
 
 f3) for the sequence of images, calculating ( 36 ) an average width as a function of the lane widths estimated at step f2) for the different images of the sequence; 
 f4) for the sequence of images, calculating ( 34 ) an average residual roll angle based on the residual roll angles calculated at step f2) for the different images of the sequence; 
 f5) updating ( 40 ):
 the roll matrix by applying a rotation according to the average residual roll angle calculated at step f4), and 
 the average reference width by the average width calculated at step f3); 
 
 f6) iterating steps f1) to f5) until the residual roll angle is lower than a predetermined threshold; and 
 f7) extracting a roll angle estimated based on the roll matrix. 
 
     
     
       3. The method of  claim 2 , wherein the step f2) of calculating the width of each of the circulation lanes and the roll angle for each couple of circulation lanes comprises:
 in case of detection in the current image of more than two circulation lanes, delivering as an output three values of lane width and two values of residual roll angle; 
 in case of detection in the current image of tow circulation lanes, delivering as an output only two values of lane width and one value of roll angle; 
 in case of detection in the current image of only one circulation lane, delivering as an output only one value of lane width. 
 
     
     
       4. The method of  claim 2 , wherein, at step f2), estimating the width of each of the circulation lanes comprises applying the relation:
     {tilde over (w)}=H   v  cos (α)( a   2   −a   1 )
 
 {tilde over (W)} being the estimate of the lane width, 
 H v  being the height of the camera with respect to the ground, 
 α being the angle of the vehicle with respect to the axis of the lane on which it runs, and 
 a 1  and a 2  being the director coefficients of the edges delimiting the circulation lane in the image. 
 
     
     
       5. The method of  claim 2 , wherein, at step f2), estimating the roll angles comprises applying the relation: 
       
         
           
             
               
                 a 
                 z 
               
               ≃ 
               
                 
                   ( 
                   
                     
                       - 
                       1 
                     
                     ± 
                     
                       
                         1 
                         + 
                         
                           W 
                           
                             
                               w 
                               ~ 
                             
                             g 
                           
                         
                         - 
                         
                           W 
                           
                             
                               w 
                               ~ 
                             
                             m 
                           
                         
                       
                     
                   
                   ) 
                 
                 × 
                 
                   
                     
                       cos 
                       ⁡ 
                       
                         ( 
                         α 
                         ) 
                       
                     
                     ⁢ 
                     
                       H 
                       v 
                     
                   
                   W 
                 
               
             
           
         
         W being the width lane reference or the estimation thereof, 
         {tilde over (w)} g  and {tilde over (w)} m  being the lane width estimates of the considered couple of lanes, 
         α being the angle of the vehicle with respect to the axis of the lane on which it runs, and 
         H v  being the height of the camera with respect to the ground.